Uncertainty-aware estimation of inverter field efficiency using Bayesian neural networks in solar photovoltaic plants

Abstract

Solar inverters are one of the most important components in a Photovoltaic plant. Their main function is to convert the DC power produced by the solar modules into AC power that can be injected into the grid. Although inverter efficiency has reached exceptionally high values, thanks to recent technological advancements, it is typically measured at dedicated laboratories under strict testing conditions, which makes its validation after deployment extremely challenging, both from a logistic and financial point of view. This paper presents a methodology for the calculation of inverter field efficiency based on Bayesian neural networks. The goal of the neural network is to model inverter efficiency and its variance as a function of the inverter's operational state. Results show that an optimised Bayesian neural network can effectively model inverter efficiency with small reconstruction errors and negligible bias. Furthermore, the model has been proven useful to replicate the calculation of the European efficiency along with a full uncertainty characterisation.

Keywords

• photovoltaic generation
• inverter efficiency
• bayesian neural networks
• machine learning
• uncertainty